SuperGPQA
SuperGPQA - graduate-level multi-domain QA benchmark.
Category
Knowledge
Score type
percent
Scale
100%
higher is better
Models tested
23
What is SuperGPQA?
SuperGPQA is a graduate-level multiple-choice exam covering 285 disciplines — a very broad, very hard knowledge test. It extends the GPQA idea to many more fields and professions.
What does it test?
SuperGPQA measures broad factual and academic knowledge. In detail, it checks:
- expert knowledge across hundreds of fields
- hard distractors with several plausible answers
- long-tail knowledge, not just famous textbook facts
- resistance to shallow pattern matching
How to read the score
The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.
Frequently asked questions
What is the SuperGPQA benchmark?
SuperGPQA is a graduate-level multiple-choice exam covering 285 disciplines — a very broad, very hard knowledge test. It extends the GPQA idea to many more fields and professions.
What does SuperGPQA test?
SuperGPQA mainly measures broad factual and academic knowledge. Concretely, it checks: expert knowledge across hundreds of fields; hard distractors with several plausible answers; long-tail knowledge, not just famous textbook facts.
How is the SuperGPQA score calculated?
The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.
Where do I get SuperGPQA data and scores?
The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/supergpqa/scores) or pull it directly from the Open LLM Radar API with any HTTP client.
Benchmark data & downloads
Where to get, inspect and reuse the data behind this benchmark.
Benchmark dataset
The question/task data itself — download and evaluate locally
Benchmark metadata (JSON)
This page's benchmark record from the Open LLM Radar API
All model scores (JSON download)
Full score table as JSON; paginate with limit/offset params
Open LLM Radar API docs
Swagger UI to explore and reuse all radar API endpoints
Model scores 23 models tested
| # | Model | Score | Date | Source | Verified |
|---|---|---|---|---|---|
| 1 |
Qwen3.5-122B-A10B
Qwen Team (Alibaba Cloud)
|
67.1%
|
01.02.2026 | Report | Verified |
| 2 |
Qwen3.6-27B
Qwen Team (Alibaba Cloud)
|
66%
|
22.04.2026 | Qwen Team (official model card) | Verified |
| 3 |
Gemma4-31B-it
Google
|
65.7%
|
16.04.2026 | Qwen Team (official model card - comparison) | Unverified |
| 4 |
Qwen3.5-27B
Qwen Team (Alibaba Cloud)
|
65.6%
|
16.04.2026 | Qwen Team (official model card - comparison) | Unverified |
| 5 |
Qwen3-235B-A22B
Qwen Team (Alibaba Cloud)
|
64.9%
|
01.02.2026 | Report | Unverified |
| 6 |
Qwen3.6-35B-A3B
Qwen Team (Alibaba Cloud)
|
64.7%
|
16.04.2026 | Qwen Team (official model card) | Verified |
| 7 |
Qwen3.5-35B-A3B
Qwen Team (Alibaba Cloud)
|
63.4%
|
16.04.2026 | Qwen Team (official model card - comparison) | Unverified |
| 8 |
Gemma4-26B-A4B
Google
|
61.4%
|
16.04.2026 | Qwen Team (official model card - comparison) | Unverified |
| 9 |
Qwen3-Next-80B-A3B-Thinking
Qwen
|
60.8%
|
10.09.2025 | Report | Verified |
| 10 |
Qwen3-30B-A3B-Thinking-2507
Qwen
|
56.8%
|
31.07.2025 | Report | Verified |
| 11 |
DeepSeek-V4-Pro
DeepSeek
|
53.9%
|
26.06.2026 | Report | Verified |
| 12 |
Qwen3-VL-30B-A3B
Qwen
|
53.1%
|
04.10.2025 | Report | Verified |
| 13 |
Qwen3.5-4B
Qwen
|
52.9%
|
01.02.2026 | Report | Verified |
| 14 |
Qwen3-30B-A3B
Qwen
|
51.8%
|
29.04.2025 | Report | Verified |
| 15 |
DeepSeek-V4-Flash
DeepSeek
|
46.5%
|
26.06.2026 | Report | Verified |
| 16 |
MiniCPM5-2B
OpenBMB
|
40.8%
|
— | Report | Verified |
| 17 |
granite-4.2-3B
IBM
|
39.9%
|
— | Report | Verified |
| 18 |
Gemma4-E4B
Google
|
38.7%
|
— | Report | Verified |
| 19 |
Qwen3.5-2B
Qwen Team (Alibaba Cloud)
|
38.6%
|
— | Report | Verified |
| 20 |
Nemotron-3-Nano-4B
NVIDIA
|
37.8%
|
— | Report | Verified |
| 21 |
LFM2.5-8B-A1B
LiquidAI
|
34.5%
|
— | Report | Verified |
| 22 |
Gemma4-E2B
Google
|
30.3%
|
— | Report | Verified |
| 23 |
LFM2.5-2.6B
LiquidAI
|
26.2%
|
— | Report | Verified |